In order to make a robot execute a given task plan more robustly we want to enable it to take care of its self-maintenance requirements during online execution of this plan. This requires the robot to know about the (internal) states of its components, constraints that restrict execution of actions and how to recover from faulty situations. The general idea is to implement a transformation process on the plans, which are specified in the agent programming language \ReadyLog{}, to be performed based on explicit qualitative temporal constraints. Afterwards, a 'guarded' execution of the transformed program results in more robust behavior.